How process understanding and business context separate real returns from expensive pilots
AI is already delivering returns in the enterprise. But the real question is where.
Developer productivity tools are working. Support chatbots handle tier-one issues. Content marketing operations, including global translations, have fallen dramatically in time and cost. These applications are deployed, measurable, and genuinely useful.
They’re also operating at the edges of what matters most to your business.
The next wave of AI value will come from applying Agentic AI to the core operational work that runs continuously: paying vendors, collecting receivables, fulfilling orders, managing inventory. Work that happens thousands of times per day and directly impacts margins, working capital, and customer experience.
The gap between current AI deployments and this next Agentic phase is simply operational context. Most enterprises are trying to deploy AI without giving it the two things it needs to work at scale: process understanding with proper observability feedback loops and business context.
Fix that, and the ROI stops being incremental.
Real Deployment, Shallow Impact
The Celonis Process Optimization Report surveyed 1,620 enterprise leaders across IT, Finance, Supply Chain, and Operations. Their findings are pretty clear: AI adoption is widespread, but depth is limited. Four in five organizations are using GenAI foundational models. Three in five have deployed chatbots for business users.
But dig into the use cases and it appears the most common applications cluster around two areas:
- Boosting developer output
- Automating tier-one support
Both are relatively easy to implement because they operate at the edges of core business operations. Furthermore, ROI from these activities is unlikely to shift the P&L in meaningful ways.
They help people work faster, but they don’t change how work flows through the organization.
AI Lacks the Map
There’s a reason most AI deployments stay shallow. Enterprises rush to implement tools and models without giving them the operational context needed to be useful beyond simple tasks.
Think about invoice processing. Looks straightforward: receive invoice, route for approval, pay vendor. Except that’s almost never how it actually works. Real invoice workflows involve multiple steps, exceptions, manual interventions, and cross-system handoffs. Some invoices need expedited approval. Others get blocked. Many require reconciliation across systems before payment can clear.
An AI without process understanding can’t navigate that. It sees “invoice received” and follows a linear script. When reality deviates (which it always does) the system stalls, creates noise, or makes mistakes that require human cleanup.
89% of leaders surveyed said it’s crucial that AI has context on how the business runs to be effectively deployed. And 58% are concerned their current processes may limit the value they can get from AI.
They’re right to be concerned. Now is the time to take it one step deeper, beyond linear SOPs fed into AI, into the age of Agentic AI.
Where the Real ROI Lives
The paper makes a straightforward claim: real enterprise value comes when AI is applied to core operational work that happens continuously, day in and day out. Activities like paying, collecting, shipping, fulfillment, procurement, etc.
The shift is from “helping people do work” to “improving how work moves.” That difference matters more than most organizations realize.
When AI assists a developer writing code, it creates local efficiency. When AI transitions to Agentic AI and optimizes an order-to-cash cycle that processes thousands of transactions daily, it creates systematic leverage. One affects individuals. The other affects margins, working capital, and customer experience at scale. We’ve seen this in our Board engagements where we dive deep on the business model and ladder-up where the AI gaps are blocking the company from true operational and profit leverage.
But to get there, AI needs two things most enterprises haven’t provided. And I see this time and time again, effectively blocking organizations ability to expand into real, durable value via Agentic AI.
The Missing Pieces
The two prerequisites are simple, but most organizations are missing both.
Process understanding means knowing how activities are sequenced across systems: what happens upstream, what happens downstream, what runs in parallel, whether tasks execute on time and in the right order. It’s a map of dependencies and a visual systems design model that can be taught to AI, in turn, so you can build Agents.
Business context means institutional knowledge codified into something AI can actually use. Things like business-specific rules, benchmarks, KPIs, etc. The logic that determines whether an order should be expedited, a vendor should be flagged, or an exception should escalate.
Without these, AI operates blind and you will never get to that Agent future everyone is yapping about. You can train the most sophisticated model on earth, but if it doesn’t understand that a partial shipment to a high-priority customer requires different handling than a routine order, it will automate the wrong thing.
Why This Becomes the Make-or-Break Layer
Unfortunately, I can’t sugarcoat this: AI deployments fail at the same rate whether you use cutting-edge models or last year’s version. The bottleneck is the lack of operational grounding.
Organizations need to understand how their processes actually run before they can expect AI, much less Agents, to improve them. That means visibility into what happens between systems, where work stalls, which exceptions occur most frequently, and how deviations from the ideal flow impact downstream outcomes. Often the best way to start is process-system diagrams (UML anyone?) that first local teams draw then expand to cross-group collaboration to see where bottlenecks are. You can even use AI to draw these and analyze them, even run simulations on how changes to stocks, flows, or multi-tier decisions affect overall COGS, operational spend, or resourcing. It’s powerful.
Process intelligence, on the other hand, creates a digital twin of operations. It connects data across systems, maps workflows as they actually execute, and makes that context accessible to AI and Agents. Instead of guessing how work should flow, Agents can see how it does flow and act accordingly.
The survey data backs this up. 81% of leaders say AI (really Agents) will be used to directly improve business processes over the next 12 months. Not to assist people in isolated tasks, but to change how core operations execute.
And when you ask Process Improvement and Operations leaders specifically, the conviction gets stronger. 89% say intelligent automation will unlock more value than anything else in the next five years. Not better analytics. Not incremental productivity gains. Automation grounded in operational understanding.
From Pilots to Process
There is a shift already happening from experimentation to integration. From asking “what can AI do?” to “where should AI act?”
The organizations moving with high impact, efficiently share a common pattern. They started with process visibility. They mapped their operations, identified where complexity creates friction, and targeted AI or Agents at the workflows that matter most to business outcomes. They gave AI the context it needs to operate safely and the process understanding it needs to deliver results.
The ones still stuck in pilot mode are usually missing one or both of those foundations.
What This Means for Enterprises
If you’re an executive evaluating AI or Agentic investments, the strategic question is really whether your organization understands its own operations well enough to make AI effective.
The easy wins are mostly claimed. Chatbots are handling tier-one support. Coding assistants are shipping with your IDE. Email copilots are writing draft responses and content operations with creative are using AI to create numerous drafts, distribute them, and translate them. These deliver value, but they won’t redefine your competitive position.
The next phase requires operational grounding. AI needs to understand how your business runs so you can graduate to a true Agentic posture. That means process visibility, business context, and the ability to act on both.
The survey data suggests we’re at an inflection point. AI is moving from peripheral productivity tools to core operational systems via Agentic technical architectures (the projects I am hands-on the most these days). The enterprises that succeed in the next phase will be the ones that solved for process understanding first.
Because AI without operational context is just expensive guesswork. And in operations that run thousands of times per day, guesswork doesn’t scale.
…
Nick Talwar is a CTO, ex-Microsoft, and a hands-on AI engineer who supports executives in navigating AI adoption. He shares insights on AI-first strategies to drive bottom-line impact.
→ Follow him on LinkedIn to catch his latest thoughts.
→ Subscribe to his free Substack for in-depth articles delivered straight to your inbox.
→ Watch the live session to see how leaders in highly regulated industries leverage AI to cut manual work and drive ROI.
Top comments (0)